Trust Me, I Am Almost a Doctor: Uncertainty in Artificial Intelligence-Mediated Doctor-Patient Interactions

The rise of generative artificial intelligence (AI) is rapidly shaping human-machine and human-human relationships. At the center of this discourse is the human-AI relationship and whether and how ethics and biases would affect such a relationship. One of the most profound fields to observe the impact of AI on such relationships is the rise of AI doctor (or “conversational AI agent”), a form of generative AI that plays the role of doctors who interact with real patients. Despite the burgeoning interest in the study of generative AI in medical contexts, the literature paints a mixed picture of its effectiveness from the lens of patients as users. The literature suggests that AI doctor has positive and negative effectiveness (e.g., accuracy, usability, user satisfaction, trust). In this study we ask: Can AI doctor respond to patients’ queries, and how does the response compare with a human doctor? To unpack this puzzle of AI effectiveness, we performed computational linguistics analysis of AI doctor’s responses to patients’ questions and how they compare with human doctor’s responses in a real-life setting using question-answer pairs data sourced from a widely used healthcare platform and AI doctor from China. Our study revealed that, overall, AI doctor’s responses are (1) more difficult to read and understand, (2) contain more similarity or duplications thus being more internally consistent. However, they contain (3) more positive emotion across several indicators than human doctor’s responses. Our findings offer insights into the uncertainty of AI doctor systems and future opportunities to advance such systems and what this means for future research on ethical AI in the medical contexts and human-AI relationships more generally. © 2025 The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.